Article
pubs.acs.org/IECR
Retrofit of Crude Units Preheating Trains: Mathematical
Programming versus Pinch Technology
Miguel Bagajewicz,*,†,‡ Gary Valtinson,‡ and DuyQuang Nguyen Thanh‡
†
University of Oklahoma, 100 E. Boyd Street, T335, Norman, Oklahoma 73072, United States
OK-Solutions, 4017 Hatterly Lane, Norman, Oklahoma 73072, United States
‡
ABSTRACT: This paper presents a comparison of two heat exchanger network retrofit methods as they are applied to crude
units: the well-known and widely used Pinch Design Method (PDM) and a recently developed Heat Integration Transportation
Model (HIT), a recently developed mathematical programming-based MILP model [Nguyen et al. Ind. Chem. Eng. Res. 2010, 49,
13]. We show that the three-step procedure (targeting, design, and evolution) used by Pinch Technology renders solutions with
excessive and unrealistic splitting of streams as well as visibly less profit compared to the results of HIT.
1. INTRODUCTION
Pinch Technology was first introduced by B. Linnhoff1,2 for
grassroots design. The method is based on a targeting phase
where the minimum utility is calculated as a function of a minimum temperature approximation, originally called minimum
approximation temperature difference (ΔTMin), later named
Heat Recovery Approximation Temperature (HRAT). The
minimum utility is therefore a floating number that depends on
the aforementioned HRAT parameter. Once the minimum
utility is fixed, one must design the network using the Pinch
Design Method (PDM)2 according to certain rules (design the
system above and below the pinch; use a tick-off rule and certain
guidelines for picking matches). Smith3 nicely presents the
method in detail.
In the case of mathematical programming, several approaches
exist. Yee and Grossmann4 provide a review of the existing
retrofit techniques in 1987. Ciric and Floudas5 proposed a
retrofit strategy using a decomposition method. Asante and Zhu6
introduced an automated approach for HEN retrofit featuring
minimal topology modifications. Finally, Briones and Kokossis7
used the hypertargets or conceptual programming approach for
retrofitting industrial heat exchanger networks. Unlike the previous approaches that are nonlinear and hard to solve globally,
our own work8 consists of a mathematical linear programming
model that economically optimizes a network by appropriate
selection of heat transferred between hot and cold intervals and
provides a globally optimal solution. This method is currently
used commercially.
Tjoe and Linnhoff9 extended the Pinch Design Method to
retrofit using targeting procedures for energy-area trade-offs
which subsequently translate into investment savings plots,
followed by a retrofit design procedure that accounts for existing
area. The method has been and is being used for years by many
consulting and engineering companies for the retrofit of heat
exchanger networks despite the high volume of mathematical
programming methods developed in the last 40 years. It has also
been heralded as the most simple and effective way to solve
problems for crude units (see for example Rossiter10). The Pinch
Technology-based retrofit method consists of three steps:9 (1) A
targeting step where energy saving are computed and capital
© XXXX American Chemical Society
investment is roughly estimated based on vertical heat transfer,
(2) a design step based on Pinch Technology grassroots design
analogies, and (3) an evolutionary step that includes a rearranging phase using loops and paths aiming at a better new area
distribution (often forced to reduce the energy savings predicted
in the targeting phase). This is also based on the grassroots
principles of Pinch Technology.
While the estimation of capital investment in grassroots design
using Pinch Technology is straightforward, this is not the case in
retrofit design. The literature is vague in agreeing on a systematic
methodology in the targeting phase. More specifically, a means to
compute precisely the amount of additional area, the number of
new exchangers and/or new shells, and the amount of repiping
required in retrofitting the network are presented much more
as a holistic procedure that requires experience in trial and error
than a systematic algorithm with concrete steps. In other words,
several alternatives are left to the discretion of the designer. Some
suggest using only the added area, but there is the possibility of
counting added shells. In addition, the method may miss the area
needed for each HRAT because it relies on the computation of
vertical heat transfer multiplied by a fixed factor called area
efficiency. The selection of this area efficiency factor is of critical
importance. The optimum is also found using different options
(net present value, payback time, etc.). It is only in the design/
retrofit phase of Pinch Technology, where details of network
configuration and exchanger areas are available, that it is possible
to compute the needed capital investment. Thus, capital
investment could be significantly miscalculated in the targeting
phase, which leads to nonoptimal solutions.
In addition, in the design/retrofit phase of Pinch Technology
as applied to crude units some specific problems arise: (i) Having
typically one cold stream versus many hot streams at the pinch,
the method requires the cold stream to be split below the pinch,
merge, and split again in a different fashion above the pinch,
rendering unrealistic designs as we shall see later in our example
Received: May 31, 2013
Revised: August 31, 2013
Accepted: September 4, 2013
A
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
same energy consumption. The target area of the retrofitted
network is therefore
for light crudes in atmospheric fractionation units. (ii) Sometimes, there is almost a second pinch, which adds additional
complexity (i.e., merging and splitting at the pinch twice). (iii)
crude units handling light crudes many times exhibit “extended
pinch” or “continuous pinch”. Although one can formally encounter a pinch point, the reality is that from the practical point
of view several values can be picked. (iv) The tick-off rule
assumes a minimum number of units above and below the pinch
separately, which leads to an excessive number of units. (v) The
number of different options to explore is sometimes daunting for
anybody that wants to explore them by hand. We will illustrate
this issue.
In this work, we show that HIT is capable of handling the
retrofit more efficiently in one single computer run, as opposed
to the trial and error pinch method. In addition, we will show that
the results are more reasonable (no splitting and merging at a
pinch), and appreciably more profitable.
⎛A ⎞
AC = ⎜ D ⎟
⎝ αr ⎠ E
C
Figure 1 also shows three different scenarios: αr = αe
(pathway 1), αr < αe (pathway 2), αr > αe (pathway 3). A
constant α pathway (pathway 1) is the most popular assumption
in retrofit studies, and while pathway 3 is desired (smaller area), it
may call for a greater degree of network modifications (area
adjustment, repiping), and hence greater capital investment.
Assuming a constant α pathway allows acquisition of AD and EC
(pinch calculations), from which AC can be readily calculated
using pathway 2. Then, the energy saving (EA − EC) can be
calculated and the capital investment can be estimated (e.g.,
purchase cost for new area can be obtained knowing AA and AC).
Finally, optimal values for energy saving and capital investment
can be found.
As one moves on a constant α pathway, the estimated capital
investment increases (more area is added) and the energy savings
increase. At some point, the investment is larger than the savings,
which indicates the optimal point has been reached. To evaluate
this tradeoff, a profitability measure (we use net present value) is
used.
It has been suggested that this procedure should only be used if
αe is greater than 0.9.11 In the event that αe is less than 0.9, instead
of equation (2), the new area ought to be calculated using the
difference of vertical areas; that is,
2. PINCH METHOD RETROFIT GUIDELINES
Retrofit design using Pinch Technology follows the same principle
as that of grassroots design: targeting first, followed by network
design. These are briefly described next:
Retrofit targeting: This task sets targets for energy saving
and capital investment that maximizes the benefit (e.g., maximal
net present value, return on investment, or minimal payback
time). Consider the energy versus area plot shown in Figure 1:
A C = AA + (AD − AB)
(3)
Retrofit Design. While retrofit targeting is quite straightforward, retrofit design is a challenging task. No method of retrofit
design has gained recognition as the “best” method. Successful
retrofit design relies on a mixture of user expertise/experience,
insightful knowledge about the process, and some inspired
guesses. Kemp12 describes three different retrofit design
methodologies. We summarize them as follows: (1) Obtain a
grassroots design for the targeted energy consumption. Where a
choice exists, favor matches which already exist in the current
network. (2) Start with the existing network and work toward the
grassroots design for the targeted energy consumption. To do
this, identify and partially remove exchangers that cross the
pinch, and complete the network so the pinch matches are
restored. If the area does not match the existing exchangers, use
loops and paths to move heat. (3) Start with the existing network
and identify the most critical changes required in the network
structure to give a substantial energy reduction. This method will
be appropriate if the “ideal” grassroots design is so different in
configuration from the existing layout that they are virtually
incompatible.
The first methodology is a bit vague and may produce
networks that do not completely use the existing area. If there is
more than one possible grassroots design (which is the case for
crude units, where there are many as we will illustrate below), this
method of enumeration may become an impractical task to
produce the optimal retrofit design.
The second methodology is a little less vague, but it requires
expertise to implement this method to do retrofit design. Even
with that expertise, it may be quite difficult to obtain the energy
target with the same area efficiency or better for a crude unit.
Examples of retrofit network design using these guidelines can be
found in Kemp.12
Figure 1. Energy versus area curves indicating (A) the existing network,
(B) the minimum area needed for energy EA, (C) the retrofit network,
and (D) the minimum area needed for energy EC.
point A refers to the existing network (which usually has some
heat crossing the pinch), while point B refers to a “minimum
area” network (obtained using vertical heat transfer; also called
“ideal network”). This network has the same energy consumption as the existing network. In addition, the “minimum
area” network points are such that there is no heat crossing the
pinch. The ratio of the minimum area network divided by
the area of the existing network is termed “area efficiency.” It is
given by
⎛A
⎞
⎛A ⎞
αe = ⎜⎜ minimum area ⎟⎟
= ⎜ A⎟
⎝ Aexisting ⎠same energy ⎝ AB ⎠ EA
(2)
(1)
The value of αe is less than 1. In Figure 1, point C refers to a
candidate retrofitted plant with reduced energy consumption
(EC). In turn, point D refers to minimum area network with the
B
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
temperature interval 20−10 °C, etc. In Figure 2, we see that from
hot stream m, interval 3, a small portion of its heat content is
transferred to cold stream n’s intervals 3 and 4. The same
happens with interval 4, which sends heat to intervals 4 and 5. In
turn, interval 5 sends heat to intervals 5, 6, and 7. One can also
note that in Figure 2 intervals 1, 2, and a portion of interval 3
for stream m, do not show any heat transferred. This is because
this heat will eventually transfer to some other cold stream. We
refer the reader to our older work13 for more explanation. The
temperature difference between hot and cold intervals has to be
positive and larger than any imposed exchanger minimum
approach temperature (EMAT).
Our contribution to this old transportation model idea is a set
of equations that can count exchangers, count shells, determine
splitting, and even consider nonisothermal mixing, all in a linear
model that does not need to consider a pinch. This is important
because in cases like crude plants the Pinch Design Method
presents problems: as we pointed out above it may exhibit a
second pinch, a continuous (or extended) pinch and, it requires
an important merge and split of the single cold stream (the
crude) at the pinch. For retrofit cases, HIT can
• add exchangers
• add shells to existing exchangers
• add area to existing shells
• relocate existing exchangers
• split or merge streams in any pattern
• perform nonisothermal mixing
• consider an exchanger minimum approach temperature
(EMAT) for each pair of streams
• avoid usage of a heat recovery approximation temperature
(HRAT)
• avoid resorting to designs “above” or “below” any pinch
temperature, which in crude units is a problem.
Moreover, unlike Pinch Technology, HIT does not rely on a
sequence of targeting followed by design. Rather, it automatically
takes into account the balance between savings and capital costs.
In addition, it can automatically generate second best, third best
solutions, etc. so it is ideal for exploring options in decreasing
order of profitability and in a systematic manner. Finally, the
model is linear, yet rigorous, which is important because it can
take advantage of existing very powerful methods. Because it is
linear, powerful (and reliable) linear solvers can be used. We use
GAMS/CPLEX.
The third methodology is not systematic and it does not
provide a recipe of actions to take (or even a list of issues to look at).
It is so ad-hoc that it needs the use of intuition.
3. MATHEMATICAL PROGRAMMING MODEL (HIT)
Our Heat Integration Transportation Model (HIT) for grassroots design and retrofit of heat exchanger networks is based on
mathematical programming. It was first developed for grassroots
design,13 and it was then extended to perform retrofit study.8 It
uses the concept of transportation of heat from hot streams to
cold streams. Indeed, the hot and cold streams are divided into
several small temperature intervals. Heat from a hot stream
interval can only be transferred to a cold stream interval if the
temperature difference allows it. The model “transports” heat
from each hot interval to possibly more than one cold interval so
that the economic objective is optimized. So far, these are the
details of an old idea (see Cerda et al.14). Indeed, heat from a hot
interval can only be transferred to a cold stream interval if the
temperature difference allows it. Figure 2 depicts a generic hot
Figure 2. Transportation model concept with hot stream m transporting
energy to cold stream n.
stream m divided into eight small temperature intervals and a
generic cold stream n divided into nine small temperature intervals. Thus, a temperature interval (e.g., temperature interval “m1”)
represents a specific temperature interval of the process streams
under consideration. For example, if the temperature span of the
hot stream is 90−10 °C (inlet T = 90 °C, outlet T = 10 °C)
then the temperature interval “m1” represents temperature
interval 90−80 °C while the temperature interval “m8” represents
Table 1. Comparison of our HIT Model and the Pinch Method
HIT model
Pinch Design method
approach
Simultaneous optimization of utility consumption and network configuration
design criterion
Simultaneous minimization of utility cost and capital investment
design principle
The whole network is designed in one single model run. Heat can transfer across the
pinch.
HIT does not use a targeting step
The whole network is retrofitted in one single model run. Heat can transfer across the
pinch. The model can compute new shells, new units, new splits, etc.
Automatically handles streams with varied heat capacity, the case of multiple utilities
and even nonisothermal mixing.
The program is user-friendly and requires no expertise from the users. The users can
control network structure by using appropriate parameter values. It can solve
industrial scale problems on a PC within a day
retrofit targeting
retrofit design
special cases
usability
C
Three-stage approach: (1) minimum utility targeting; (2)
network design for minimum number of units; (3) use
loops and paths to evolve.
Same criterion but in a hierarchical manner: optimize utility
consumption first, then the network configuration
Design starts from the pinch point, sections above and below
the pinch are designed separately.
Performed using an estimation of the cost of total area added.
Typically the grassroots design is used as a starting point.
Need special treatments for special cases like streams with
varied heat capacity, multiple utilities or multiple pinches
Require expertise of the user to make all the right decisions on
splitting of stream, etc. according to the pinch design rules.
The design process is time-consuming, especially for large
scale problems and presents multiple alternatives.
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 3. Existing network design for a crude at 31.8 API and 10 hot streams.
Table 2. Exchanger Information for Example
In our second paper,8 we describe the extension of the
grassroots model to retrofit. We presented two models, one that
considers adding shells to existing and new units and another that
also considers unit relocation.
To summarize the differences between the HIT model and the
Pinch Design Method, HIT requires a single model run to optimize
the utility consumption and network design (configuration, shells,
units, splits, etc.) as per provided economic parameters. It does not
require a targeting step and does not have heat transfer restrictions
associated to a pinch. A versatile approach, HIT does not have
difficulty handling special circumstances like varying heat capacities,
multiple utilities, nonisothermal mixing, or phenomena in the
composite curves. HIT only requires implementation of network
parameters to the model, thus it is very user-friendly. In turn, the
Pinch Method requires a targeting step to determine the optimal
utility followed by designing a network. Design starts at the pinch
with a network developed separately above and below the pinch.
Even when retrofitting, the grassroots design is typically the starting
point. The method requires special treatment for special circumstances like those listed above. It is not a very user-friendly method
and requires expertise to make the right decisions. It can also be very
time-consuming if there are many possibilities for the network.
D
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Table 3. Heater/Cooler’s information for Example
Table 4. Overall Heat Transfer Coefficients and Areas for Exchangers in Example
U (kW/(m2·K))
area (m2)
EX1
EX2
EX3
EX4
EX5
EX6
EX7
EX8
EX9
EX10
EX11
0.2
707
0.4
63
0.3
182
0.35
96
0.2
411
0.2
240
0.4
74
0.35
68
0.2
473
0.28
324
0.31
154
The characteristics of the HIT model, as compared against the
pinch design method, are summarized in Table 1.
Table 5. Assumptions Made for This Analysis
furnace utility costs ($/(Kw·yr))
area addition costs ($/m2)
fixed exchanger cost ($)
maximum area per shell (m2)
installation cost as a percent of equipment cost
interest rate
furnace efficiency
4. EXAMPLE
Consider an existing network (Figure 3), which uses one crude
oil with an API of 31.8 at 8330 ton/day. Currently, the furnace
heats the crude from 234.8 to 350 °C, with a furnace utility of
39.09 MW. Table 2 provides exchanger information for the flow
rates of each stream, the temperatures of the streams in and
out of the exchanger, and the duty of each exchanger. Table 3
provides the same information for the heaters and coolers.
Table 4 provides the overall heat transfer coefficients and areas of
each exchanger.
Assumptions. In developing and economically analyzing
the Pinch Technology and HIT networks, we required a
few assumptions regarding the design decisions and economic
parameters. First, we assumed that exchangers did not have to
stay in a particular order. If the model called for a change then
we permitted it without considering additional piping costs.
Although alternative heat exchangers may be applicable to a
given network, and probably have a higher benefit, this study is
only looking at the use of standard S&T heat exchanger units. We
priced the exchangers on the basis of a fixed cost per shell, a
variable cost per area, and installation costs at 50% of the
equipment costs. The maximum area for a given shell is assumed
to be 500 m2. We assumed the cost of energy to heat the furnace
is $100 per kW-year. The furnace runs at 80% efficiency. We only
consider the exchanger investment and the energy savings. Costs
due to splitting and cooling are not considered in the capital costs
as they have a much smaller impact. Table 5 summarizes all of our
economic assumptions.
We also develop simulations matching measured values using
PRO/II (Invensys) to see how the network actually performs.
This is essential as a simulation can show possible problems with
both model results. This is so because, especially for Pinch
Technology, assumptions of average heat capacity are usually
taken.
Using the defined data from this simulation, composite curves
are constructed. These composite curves are developed using the
average FCP values across a given stream and the initial and final
temperatures of each stream. The crude stream is divided into
100.00
271.20
127,129
500
50%
10%
80%
Figure 4. Example network composite curves showing energy crossing
the pinch at 210 °C.
multiple intervals after the desalter to account for FCP changes,
which gives rise to the curved cold composite. We also note
that no vaporization occurs. Figure 4 shows the composite
curve for the example network. The HRAT selected for these
curves is the minimum temperature difference between the hot
and cold side of all exchangers in the existing network (21 °C
at exchanger EX3). We also show the amount of heat crossing
the pinch.
The pinch occurs at 210 °C for the hot composite curve. At
this HRAT, a minimum utility of 24.82 MW is possible. No
problematic phenomena such as a second pinch or continuous/
extended pinch occur in this example, meaning that Pinch
Technology should have no trouble producing a network. The
network has 15.73 MW crossing the pinch, something that needs
E
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
fixing. This is the difference between the current furnace utility
and the minimum utility at an HRAT of 21 °C. In the network, it
relates to the exchangers that experience pinch crossing: EX5,
EX7, EX8, and EX9. It also includes any energy above the pinch
that dumped into the cold utility: SR-Q and VR2.
Pinch Technology Retrofit Targeting. For retrofit
targeting using Pinch Technology, we sought the optimal net
present value of savings over 10 years minus the investment
(additional cost of new exchangers and new shells for existing
exchangers). It is not clearly stated in the different pinch retrofit
procedures that fixed costs per shell ought to be added, but
because HIT actually considered them, we added to the cost per
area the cost of new shells. To obtain the optimal NPV, we varied
the utility usage based on an area efficiency of 0.68, the existing
network value. As this value is smaller than 0.9, we used eq 3 to
calculate the targeted retrofit area. When the piecewise FCP is
used for the crude, the maximum NPV is obtained for HRAT =
30 °C (see Figure 5), which has a minimum utility of 27 MW with
Figure 6. Comparison of composite curves for average and piecewise
FCP.
Figure 7. Results from the targeting phase indicating the payback for
α = 1 (suggested when α < 0.9) and α = 0.68 (current network efficiency).
Figure 5. Results from the targeting phase (NPV) for average and
piecewise FCPs.
Similarly, if we use paybacks of 2 or 3 years (see Figure 7), for
the piecewise FCP case only, which is the one we use from now
on, the answer is different. In the same figure, we also provide a
comparison of the savings versus the investment if the current
efficiency of 0.68 was used in equation 2.
Pinch Technology Network Design. We start using the
NPV criterion, by identifying which exchangers cross the pinch
(Figure 8). As we indicated earlier, no vaporization occurs for the
cold streams before the furnace. We still use piecewise FCPs for
the crude after the desalter to account for adjustments in FCP
in our analysis. In Figure 8, we mark all exchangers that cross
the pinch and indicate the energy crossing the pinch with
parentheses. The FCP of the crude stream into the furnace ranges
between 224 and 328 kW per °C. We indicate only the average FCP
of 288 kW per degree °C for the crude after the desalter.
Above the pinch, there is only one possible pinch design, but
below the pinch, the number of different networks is too large for
any pinch expert to perform by hand in a reasonable amount of
time. Indeed, we found that 23 possible networks are possible
using only 11 exchangers below the pinch (the predicted
minimum number). Of the 23, only one solution uses all the
exchangers present in the original network and does not add new
matches before the desalter (Figure 9). In Figure 9, dotted lines
indicate new exchangers and solid lines indicate existing
exchangers. If new area is to be added, a +A sign is located
close to the exchanger. All splits are new.
the pinch remaining at 210 °C. This amounts to a total savings of
12.09 MW of energy. The discontinuities in the NPV curves
correspond to changes in the number of shells. Figure 5 also
shows the NPV curve corresponding to retrofit targeting using
average FCP. Notice how the NPV curved based on average FCP
shows a much more optimistic NPV with the optimum to be
around a hot utility value of 23.5 MW (this corresponds to an
HRAT = 28 °C). This difference arises because the average FCP
of the cold composite underestimates the minimum utility. The
cold composite consists of two streams, a stream before the
desalter and a stream after the desalter. The average FCP of the
stream after the desalter includes the energy before the pinch and
after the pinch. Thus, using the average FCP above the pinch
underestimates the minimum utility and overestimates the
energy required to reach the pinch. Figure 6 shows the composite
curves for each case, that is, piecewise and average FCP. The area
below the pinch increases when using average FCP, but it also
decreases above the pinch. In addition, we found that the total
area does not differ significantly between the two cases. The
minimum utility of the average FCP case for HRAT = 30 °C is
approximately 24 MW, rendering a difference between the two
methods of approximately 3 MW. As we want to utilize pinch in a
way that obtains the best network possible, we base our analysis
on the piecewise retrofit targeting.
F
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 8. Original network. Energy crossing the pinch is in parentheses and is represented by shaded exchangers; stream connections represent existing
exchangers.
Figure 9. Pinch Design Network: Solid connections are existing exchangers, dash connections are new exchangers, +A represents area addition.
below the pinch and 8 above (the minimum possible) exhibit this
problem of splitting and remerging at the pinch. All other splits
are made to keep the network at an HRAT of 30 °C.
As per the network before the desalter, it also exhibits
an undesired splitting complexity that if explored further may
Once we designed the network of Figure 9, we found that the
number of splits is characteristically high in addition to the
problematic merging and resplitting at the pinch (the pinch is
located after the exchangers in two branches after the desalter,
namely MCR and HVGO). All networks with 11 exchangers
G
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 10. Pinch design network detailing loops: (a) loop starts at exchanger 4 and proceeds as follows 4−5−11−12−4; (b) loop starts at exchanger 6
and proceeds as follows 6−8−6.
Figure 11. Pinch Design Network after loops: solid connections are existing exchangers, dash connections are new exchangers, +A represents area
addition.
be simplified. Other networks of the 23 found could exhibit
more simplicity (more coolers with fewer splits). We omit
discussing all the alternatives as they share the same complexity
problem.
Pinch Technology Evolution. The next step suggested in
the Pinch Design Method is to use loops to simplify the network,
mostly to reduce the area addition to existing exchangers and
replace it with area added to new exchangers. After this, one can
use paths to restore HRAT if needed and/or desired. The
network has 22 observable loops. Figure 10 indicates the specific
loops used to transfer area. The resulting network shown in
Figure 11 is the PDM network design after loops only. The
network is feasible with an EMAT = 20 °C. We notice that aside
from transferring area, one major concern is to reduce the
splitting, which was partially accomplished.
Figure 12 depicts the paths used to restore the retrofit target of
HRAT = 30 °C. We determined that there are five possible paths
in the network. If the minimum temperature difference in the
network is reasonable (which it is with an EMAT = 20 °C), it may
be advisable to keep the network as is with an EMAT smaller than
the HRAT selected and not use a path. If the reduction in energy
does not reduce a sufficient amount of area, then the network
in Figure 11 would be more desirable. The optimal trade-off can
be determined through economic analysis of the networks.
Figure 13 shows the network after paths have been explored. Any
further changes would be allowing for pinch crossing methods or
knowledgeable guesses.
Although the network of Figure 13 improves the area slightly,
the economics will show that it is not enough to overcome the
reduction in savings of 1 MW. Thus, the best network design
H
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 12. Pinch design network detailing paths: (a) path starts at hot utility and proceeds as follows H-4−11−13-C; (b) path starts at hot utility and
proceeds as follows H-7−9−13-C.
Figure 13. Pinch Design Network after loops and paths: solid connections are existing exchangers, dash connections are new exchangers, +A represents
area addition.
corresponds to Figure 11 where an EMAT of 20 °C is used. The
network uses 5889 m2 across 13 exchangers requiring the addition of only 13 shells. Its furnace utility is 27 MW with an outlet
temperature of 279.8 °C. A process flow diagram of Figure 11 is
provided in Figure 14.
Something to consider is that the network area of the pinch
retrofit design does not completely use the existing area. The
only way of fixing this is to break the tick-off rule and allow the
energy to be shifted around so that the existing area is completely
used. The existing network has pre-existing coolers that would
allow for this. The only other issue that we would like to examine
further with pinch is the number of splits.
Analysis beyond Pinch Technology. While pinch retrofit
recipes suggest that loops and paths be used to move new area
around, preferably to a new exchanger, the network may still
remain overly complex due to splits. The pinch design for our
example has too many splits, a total of five with one using
multiple branches. To reduce the number of splits while keeping
the same utility usage, we took the network after loops and then
looked at the possibility of allowing some matches to crisscross
the pinch. We considered the HRAT to be fixed; therefore, the
minimum utility would not be reduced, and we used an EMAT
of 20 °C.
Through analysis of the location of each exchanger, we
determined that the Crude2−LGO match could be before the
Crude2−HVGO match. We then removed the split after the
pinch and moved the Crude2−HGO match onto another
branch. This resulted in a simple structure around the pinch but
with issues in the temperature difference. By reusing the loops
and adjusting the FCP of the crude between the branches,
I
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 14. Pinch Network process flow diagram: design based on the pinch design network after loops (see Figure 11).
Figure 15. Pinch Design network modified for reduced splitting: solid connections are existing exchangers, dash connections are new exchangers, +A
represents area addition.
the network EMAT was restored to 20 °C. With the allowable
temperature difference that low, we also removed some of the
splits away from the pinch. In particular, we removed the second
split before the desalter and placed the Crude1−LGO match in
series with the Crude1−HGO match. Energy was transferred around
a loop once again to keep temperature differences above 20 °C.
J
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 16. Modified Pinch Network process flow diagram: design based on reduction of splits by allowing crisscrossing at the pinch (see Figure 15).
Figure 17. HIT retrofit design: solid connections are existing exchangers, dash connections are new exchangers, +A represents area addition.
HIT Results. As indicated, HIT uses temperature intervals of
heat transfer to determine the design, not the pinch. All that we
obtained from Pinch Technology is the minimum utility at
various HRATs. The method can also have constraints added,
like for instance, the maximum number of splits for each individual stream. With these details, the software determines the
Finally, the last three exchangers are placed in series. The temperature difference was adequate so no paths were required to keep
above the EMAT of 20 °C. This produces a far simpler network with
only one split before the desalter and one split after the desalter with
three branches. The resulting design detailing the matches is depicted
in Figure 15, the network process flow diagram is in Figure 16.
K
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Table 6. Comparison of HIT and Pinch Technology Results and Simulated Result
HIT
new area (m2)
new shells
furnace utility
(MW)
result
simulation
result
4077.4
12
25.96
3881.8
13
25.98
pinch
error
result using average
FCP
result using piecewise
FCP
result from
simulation
error for average
FCP
error for piecewise
FCP
−5.0%
7.7%
0.1%
3152.2
15
26.32
3791.8
14
26.96
3638.6
14
27.0
13.4%
−7.1%
2.5%
−4.2%
0.0%
0.1%
Figure 18. HIT network process flow diagram: based on HIT retrofit result.
possible and it would reduce the error at a larger computational cost, but as we show later, the difference in NPV is small
enough to accept the result. To compare HIT to pinch, if we
obtained the area using the average FCPs and temperatures
in pinch for the initial pinch result (Figure 8), we would find
that it underestimates the area by 13.4% compared to the
simulation of the pinch network. In addition, it overestimates
the possible energy savings. In turn, the deviations are smaller
when the piecewise FCP is used. The disparities between the
expected result and simulated results for HIT and Pinch are
provided in Table 6.
The HIT design of Figure 17 is simple, with only one split
before the desalter and one split with three branches after the
desalter. The network reaches a furnace utility of 25.96 MW, with
a furnace inlet temperature of 283.6 °C. The required network
area is approximately 6627 m2 across 14 heat exchangers. This
area required only 13 new shells across the network. This means
that as far as design goes, the network is much more feasible, but
uses roughly 750 m2 more and the same number of shells as the
Pinch Technology design. The HIT network has 1 MW of
savings more than the Pinch Technology network. Figure 17
shows the HIT design using matches while Figure 18 shows the
HIT design process flow diagram.
optimal network that meets the constraints and uses the optimal
area and heating utility. While costs of splitting are not considered (as in the case of Pinch Technology), we did limit the
network to only one split before the desalter and three splits after.
Although it is possible to restrict the location of a particular
match, we did not place this restriction on our HIT run.
We ran HIT with EMAT > 5 °C as a constraint (no limitations
on the utility used), maximizing NPV, and it produced the
network seen in Figure 17. We then simulated the network using
only the locations and duties of each exchanger obtained by the
software results. Although the split flow rates are also provided by
HIT, we usually see if adjusting the flow rates can minimize the
area further in the simulation, just as we did with the results of
Pinch Technology. The area determined by the software is not
without error, the severity of which is dependent on the number
of temperature intervals used and the FCP associated to each
interval. For this network, we found that using 140 temperature
intervals in HIT overestimated the area by 5%. It should be noted
that we only used 52 different FCP values (25 FCPs for the 11
hot streams, 25 FCPs for the 2 cold streams, and 1 FCP each for
the hot and cold utility) across all 140 intervals, which is where
most of the error may arise. HIT predicted one less shell than
what simulation indicates is required. Using more intervals is
L
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
The HIT network has fewer splits and branches than the pinch
design. It is more comparable to the design after analysis beyond
PT. This HIT network is a simple network with higher savings
and comparable area requirements.
The economics are detailed in Figure 19, showing a comparison of the net present values of all the solutions. First, we
Table 7. Summary of HIT and Pinch Results
energy savings (MW)
new area (m2)
no. of new shells
no. of new units
savings ($ million/year)
investment ($ million)
NPV ($ million)
HIT
pinch after
loops
pinch after
reducing splits
25.96
3881.8
13
3
$1.64
$4.06
$6.02
27.0
3722.8
13
2
$1.51
$4.00
$5.29
27.0
4258.4
15
2
$1.51
$4.59
$4.69
possible to make the pinch retrofit method succeed in proving its
best. We even went farther from its recommendations to simplify
the network, one of the biggest problems. Indeed, pinch renders
two splits before the desalter and three splits after the desalter
with as many as four branches when only loops and paths were
considered. The HIT network, in contrast, requires one split
before and one after the desalter. Finally, HIT obtains an NPV
over 10 years of $6.02 million, while pinch does not exceed an
NPV $5.29 million.
5. CONCLUSIONS
We pointed out the difficulties one can have using the pinch
design method in crude units. These difficulties are rooted in the
fact that it requires starting the designs at the pinch, which leads
to branching above and below the pinch utilizing a lot more
exchangers than needed, and especial expertise on the subject to
“evolve” the design. Our HIT model provides more reasonable
and more profitable designs, without the need of user built-in
expertise in one single run. Specifically:
• The strict pinch retrofit result is overly complex from the
operational point of view. Even when loops are used to
simplify it, the network is still complex, albeit a bit better
economically. Usage of paths makes matters worse.
• HIT predicts a network with larger savings (1 MW more)
than the pinch targeting predicts. This has an impact on
the NPV. The network also exhibits a more reasonable
crude split.
• HIT largely outperforms the pinch result rendering a
simpler network and a higher NPV (∼15% better in this
example).
• The roots of the problems for Pinch Technology are (a)
bad targeting (pessimistic in this case; 1 MW difference
with HIT) and, (b) complex networks, that is, too many
splits.
• After one attempts to fix the splitting problems using
analysis beyond pinch technology, the economics actually
deteriorates further.
Finally, we point out the need for simulations (we used Pro II)
to provide a realistic analysis of the networks. Because of these
simulations, both HIT and Pinch Technology exhibit adjustments
in area calculations, and the economics is adjusted accordingly.
Figure 19. Economics of HIT and Pinch Technology indicating the
NPV for the HIT result: all points in the development of the pinch
network, and the corresponding simulations.
note that pinch targeting (the curve on top) is very optimistic
because it underestimates the area needed. Once the network
was simulated, we calculated the areas and obtained the point
labeled “Pinch Retrofit Result 1 Simulated”. The network
corresponding to Figure 11 (labeled “Pinch Retrofit Result 2
after loops”) has the best NPV for Pinch Technology. The use of
paths to restore the minimum temperature difference, as in
Figure 13, results in a poor NPV.
A few critical details can be noticed from Figure 19. First, the
use of average FCP can lead to results that are not realistic. This
disparity is indicated by the smaller hot utility and larger NPV
when using average FCP compared to piecewise or simulated
pinch results. The NPV is larger because of the difference in area
from the network and minimum utility. Second, the pinch
simulated result is close to the piecewise calculation, but for the
use of average FCP the result is quite different. The simulated
design of pinch performs better than the piecewise design
because the simulation has very small intervals of FCP. The use
of average FCP gets both the furnace utility and the area wrong as
indicated earlier. The HIT result performs slightly better than the
HIT result simulated. This is an issue with the number, size, and
FCP of intervals used. In this case, HIT did not see the area that
needed to be added to one of the exchangers, while simulation
indicates that a new area, and thus a new shell, must be added.
Table 7 summarizes our findings for the HIT result and the
pinch result for the best pinch network and a simplified pinch
network. We notice that the results from HIT and from Pinch
Technology differ from their corresponding simulated results.
Simulations are done in such a way that the energy consumption
is the same. Thus, the area is adjusted and therefore the NPV
changes a little. The results of this analysis indicate that HIT
produces (a) a more reasonable network, and (b) a more
profitable network than pinch technology.
HIT found an optimal furnace utility (26 MW) smaller than
the utility retrofit targeting predicts (27 MW). We made all effort
■
AUTHOR INFORMATION
Corresponding Author
*E-mail: bagajewicz@ou.edu.
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS
To perform our simulations we used Pro/II from Simsci/Invensys.
M
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
■
Article
REFERENCES
(1) Flower, J. R.; Linnhoff, B. A thermodynamic−combinatorial
approach to the design of optimum heat exchanger networks. AIChE J.
1980, 26 (1), 1−9.
(2) Linnhoff, B.; HindMarsh, E. The pinch design method for heat
exchanger networks. Chem. Eng. Sci. 1983, 38 (5), 745−763.
(3) Smith R. Chemical Process Design and Integration; McGraw Hill:
New York, 2005.
(4) Yee, T. F.; Grossmann I. E. Optimization model for structural
modifications in the retrofit of heat exchanger networks. Proceedings of
the First International Conference on FOCAPO; Elsevier Science: New
York, 1987; pp 653−662.
(5) Ciric, A. R.; Floudas, C. A. A mixed integer nonlinear programming
model for retrofitting heat-exchanger networks. Ind. Eng. Chem. Res.
1990, 29, 239−251.
(6) Asante, N. D. K.; Zhu, X. X. An automated approach for heat
exchanger network retrofit featuring minimal topology modifications.
Comput. Chem. Eng. 1996, 20 (Suppl), S7−S12.
(7) Briones, V.; Kokossis, A. C. Hypertargets: A conceptual
programming approach for the optimization of industrial heat exchanger
networksII. Retrofit design. Chem. Eng. Sci. 1999, 54, 541−561.
(8) Nguyen, D. Q.; Barbaro, A.; Vipanurat, N.; Bagajewicz, M. J. All-atonce and step-wise detailed retrofit of heat exchanger networks using an
MILP Model. Ind. Eng. Chem. Res. 2010, 49 (13), 6080−6103.
(9) Tjoe, T. N.; Linnhoff, B. Using pinch technology for process
retrofit. Chem. Eng. 1986, 93 (8), 47−60.
(10) Rossiter, A. P. Improve energy efficiency via heat integration.
Chem. Eng. Prog. 2010, 106 (12), 33−42.
(11) Shenoy, U. V. Heat Exchanger Network Synthesis: Process
Optimization by Energy and Resource Analysis; Gulf Publishing Company:
Houston, TX, 1995.
(12) Kemp, I. C. Pinch Analysis and Process Integration: A User Guide on
Process Integration for the Efficient Use of Energy, 2nd ed.; ButterworthHeinemann: Boston, MA, 2007.
(13) Barbaro, A.; Bagajewicz, M. New rigorous one-step MILP
formulation for heat exchanger network synthesis. Comput. Chem. Eng.
2005, 29 (9), 1945−1976.
(14) Cerda, J.; Westerberg, A. W.; Mason, D.; Linnhoff, B. Minimum
utility usage in heat exchanger network synthesis. A transportation
problem. Chem. Eng. Sci. 1983, 38 (3), 373−387.
N
dx.doi.org/10.1021/ie401675k | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX